AI-built businesses, agents and tools that are really earning, found by reading public records, payment data and live products. Each one comes with how we know.
Ordered by the money we could see, biggest first. Figures we checked ourselves come before ones someone else claims, and finds with no figure come last. The summary is free; each find's full story is for subscribers.
Showing 109 of 109 entries.
Earners. Ordered by the money we could see, biggest first. Figures we checked ourselves come before ones someone else claims, and finds with no figure come last.
A service that sells lists of estates going through probate to property investors reports 68.6k dollars of monthly recurring revenue from 617 subscriptions.
A tool that automates WhatsApp messages for online shops, founded in January 2026, reports 26.0 thousand dollars of monthly recurring revenue from 249 subscriptions.
A booking and client-management tool for therapists reports 20.9k dollars of monthly recurring revenue from 370 subscriptions, and 1.15M since it began in 2020.
An AI-native security team has been paid 198,983 dollars for valid bug reports on Immunefi. Only 11.5k came in the last three months, so the pace has cooled.
A tool that lets small businesses build their own AI support chatbot without code reports 17.2k dollars of monthly recurring revenue from 301 subscriptions.
In Shopify's app store, 64% of random AI-named apps have no reviews at all. The survivors are plain utilities, such as an alt-text generator with 180 reviews.
A website that colours old black-and-white photos took 3,912 dollars in 30 days at 1 dollar a photo, while the open-source model behind it costs 0.042 dollars a run.
A web service that sells answers one request at a time took $1,781.84 over 30 days from 28,971 calls and 121 buyers. More than half came from one buyer.
An AI tutor for Indian accountancy students gives guests 15,000 free tokens, then sells a 299-rupee monthly pass. It cites official course books, which is the copyright worry.
This pay-per-call service took in 398.67 dollars over a month from 734 requests. Nearly all of it came from a single buyer, so the picture is thinner than it looks.
A young business sells live-interview AI help to job seekers from 449 rupees a session, and an AI first-round interviewer to employers. The job-seeker side is the shaky one.
A service whose name says it referees disputes between software agents took $187.28 in 30 days from 1,018 calls. Three buyers paid, almost all from one.
A pay-per-call service earned 176.25 dollars in a month, but spread across 643 different buyers. That works out at pennies each, the opposite of a one-big-customer story.
A service that searches records about software agents took $157.70 in 30 days from 5,329 calls and 28 buyers. Many small payments and no dominant buyer, but a small total.
A small pay-per-request gateway took in $133.03 in a month. Nearly all of it came from one buyer, so the real lesson is how fragile a single customer can be.
An AI go-to-market platform reports 424k dollars of monthly recurring revenue and 1.80M all-time, but the revenue-tracking link expired and last synced on 14 August.
Australian shares seem to drop slightly less than the dividend they pay, so buying just before payout has averaged 0.62 percent a trade since 2015. But the edge is shrinking.
Atlassian's Forge marketplace takes no commission up to 1M and shows install counts. But in most niches an established leader already sits on top, and most apps get few reviews.
After directors buy their own company's shares, those shares beat random dates by about 2.6 percent over 60 days. After costs, and against small companies in general, little is left.
Fees brought in 8-17% of capital a year on a fixed-range Uniswap pool, but holding the coins beat it in two of four years because price kept leaving the range.
Projects pay researchers who find flaws in smart contracts, but duplicates are the main loss and bots already catch the easy bugs, so generic AI output tends to earn nothing.
One US company runs a growing set of local-trade websites, each with an AI chat assistant taking job requests. Growth past 30 sites, or real traffic, would suggest it pays.
Among 3,073 Polymarket wallets, traders differ stably from their peers, but after correcting for luck only 1 wallet still stood out. Profit rankings mostly show big accounts.
Groups of near-identical sites reveal what operators are betting on. The research found 8 groups of 5 or more sites, but most appear in one burst and then stop growing.
Niche sponsors put up prize pools of 5k-1M to attract developers, and few people enter. A prize can be large per entrant, but it is top-heavy: most entries win nothing.
One freelancer shipped 5 client chatbots in 13 months for high-ticket businesses like aged care. New clients arrived every month or two, with almost no web presence.
Of 170 public chatbots checked, 94 no longer worked. Most were hobby projects, so whether local businesses would pay someone to keep their bots running is still an open question.
Of 392 small web tools whose founders share real revenue, 51.5 percent earn anything. The money is lumpy: one business holds 90 percent of the sample's total.
One Australian company runs 3,456 local cost-guide pages and sells job leads from them. Its own page shows just 6 house-cleaning jobs nationwide in 30 days.
Nine wallets selling data to AI agents took in under 1 USD from independent buyers in about 2.8 days. A Korean public-data service shows the idea works in theory.
In a few days, 13,831 AI agents registered on one public list, yet only about 0.14 percent had a distinct, working service. Real outside buyers look rare.
A doctor's guide to being found by AI shows '363/450 copies sold' with 87 left. The count stayed identical across a short watch, so the number may be decoration.
On eight prediction-market sites, contracts priced at 95-99.9 cents fail about as often as the price implies. After fees there is little or nothing left. Information, not advice.
Only 20 percent of slots on Bittensor's AI competitions earn anything, though miners overall collect 2,811 TAO a day. The token price decides what that is worth.
A marketplace where AI agents buy tool calls from each other handled 11.97 million requests for 24,533 dollars in fees. 74 percent came from one group of trading agents.
One security-contest platform has paid out 14.3 million dollars in total. The top five contestants hold 21 percent, each with 408 to 933 thousand dollars.
A public dashboard showed 90,148 dollars of fees in 30 days for an agent-services platform, but the payments were near-uniform amounts of 0.073 dollars and look recycled.
An Estonian game publisher filed 3.80 million euros of revenue and 2.83 million euros of net profit for 2024, with 99 percent of the revenue from the US.
Bankr, a launchpad for AI trading agents, shows 32,779 dollars of fees in 7 days and 493,671 in 30 days, but fees are down about 95 percent from September.
Hipages, an Australian marketplace that sells leads to tradespeople, reported 83.1 million Australian dollars of revenue for the year, 97 percent of it recurring, at a 24 percent margin.
Chatbase, a builder of AI support agents, says in a payment-processor case study that it reached 10 million dollars of annual recurring revenue with 26 staff and about 10,000 businesses.
Clanker, a tool that launches tokens, shows 1.35 million dollars of creator fees in 30 days, with monthly totals that swing between 0.3 and 3.3 million.
A tool that lets crypto users cancel risky permissions took 2,279 dollars from 21 payers on one network and 3,565 dollars from 10 payers on another in 30 days.
Virtuals, a platform that launches AI agents and runs an agent marketplace, shows 13,363 dollars of protocol revenue over 7 days and 105,964 over 30 days.
XBOW, a company that automates security testing with AI, has about 1,060 reports on the HackerOne bounty platform and under 40 thousand dollars of bounties from them.
A tool that manages people's liquidity-pool positions automatically managed 538,804 dollars of fees in 30 days and keeps an 8 percent commission on them.
One new crypto app fell from 22.4 million dollars a month to 2.2 million, another from 407 thousand to 6 thousand, a third from 987 thousand to 3.2 thousand.
A business-for-sale listing says an AI song-gift shop took 2.66 million dollars of revenue and 664 thousand of profit in 8 months from 126,752 orders averaging 21 dollars.
Three shops that sell downloadable templates on a handmade-goods marketplace show sales counters of 177.5K, 214.6K and 158.6K, at median prices of 24.00, 7.20 and 7.36 dollars.